COVID-19 and High School Attendance: Modeling Latent Trajectories and the Influence of School Policies and Practices

Abstract

Objective:

School attendance is important for adolescent and adult health but drastically decreased during the COVID-19 pandemic. We aimed to identify distinct attendance trajectories across a 5-year period spanning the onset of the COVID-19 pandemic and examine the influence of school policies and practices related to school discipline and college readiness programming on these trajectories.

Methods:

We analyzed administrative data from the 2016–2017 to 2020–2021 school years for 2,805 rising high school students (7^th or 8^th grade at baseline) at 5 Southern California schools participating in a randomized trial of the AVID college readiness program. We identified attendance trajectories using group-based trajectory modeling then estimated the influence of suspension and AVID participation by including these as time-varying covariates.

Results:

We identified three attendance trajectory groups: stable high attendance (77.2%); acutely declining attendance (17.7%); chronically declining attendance (5.1%). Suspension was associated with decreasing attendance across all groups (stable high: β= −8.96; p < 0.0001; acutely declining: β= −12.00; p < 0.0001; chronically declining: β= −7.61; p = 0.025). AVID participation was associated with a small decrease in attendance in the stable high attendance group (β= −0.97; p = 0.025) and a larger increase in the acutely declining attendance group (β= 3.60; p = 0.015).

Conclusions:

These findings increase our understanding of the impacts of the COVID-19 pandemic on adolescents’ school attendance. The health and educational impacts of the COVID-19 pandemic are not uniformly severe, and both tailored individual and universal school-level interventions are likely needed to support all youth to thrive.

Attending school is crucial for educational success and health.¹ Consistent attendance offers opportunities to connect with the school community and benefit childhood health through relationships with supportive school-based adults and healthier peer networks.^2–7 Attendance is strongly associated with educational attainment, including high school graduation,⁸ which is a critical social determinant of health.^1,9 Higher educational attainment is associated with decreased rates of chronic conditions including cardiovascular disease and depression^10,11 and longer life expectancy.¹²

Although attendance is critically important, it is also uniquely challenging.¹ To optimally attend school, youth must be physically and emotionally well, have access to health care needed to stay this way, and have their social needs met, including stable housing, adequate nutrition, and reliable transportation.^1,13–15 Additionally, a sense of school connectedness, the feeling of being cared for, valued, and belonging at school, is an important driver of attendance.⁷ School staff, policies, and practices all have roles in promoting attendance by fostering high-quality school environments that are safe, supportive, and engaging for all learners.³ Overall, attendance is a unique holistic marker of youth well-being with relevance to clinicians and educators alike.^1,16

The state of attendance worsened in the wake of the COVID-19 pandemic. Data collected in the 2017–2018 school year indicated that 8 million or nearly 16% of U.S. students were chronically absent, defined as missing at least 10% of school days.¹⁷ By the 2022–2023 school year, nearly 30% of U.S. students were chronically absent.¹⁷ Considering these findings, concerted efforts are needed to comprehensively support youth and enhance education and health outcomes. Increasing our understanding of attendance during the COVID-19 pandemic can support these goals.

Attendance generally declines across school transitions (e.g., middle to high school) and throughout the high school years¹⁸ but not uniformly for all students.^19–21 Students experiencing greater disruption in school connectedness across school transitions have higher likelihood of shifting to more problematic attendance trajectories over time, a phenomenon that may have also occurred across shifts to remote instruction during COVID-19.¹⁹ School policies and practices may similarly impact attendance by building supportive connections (e.g., fostering academically-motivated peer groups^3,7) or breaking them (e.g., utilizing exclusionary discipline strategies including suspension and expulsion²²). Understanding the shapes of and influences on attendance trajectories during the COVID-19 pandemic may support identification of youth with unique education and health needs and inform intervention.

We aimed to identify distinct attendance trajectories between the 2016–2017 and 2020–2021 school years among a sample of youth enrolled in 5 public high schools in Southern California at the onset of the COVID-19 pandemic. We used group-based trajectory models (GBTMs), statistical models used to classify subgroups of individuals following similar patterns over time, to characterize attendance trajectories. We further estimated the influence of two school policies and practices perceived to impact school connectedness—school suspension and college readiness programming including a mentoring component—on the shape of attendance trajectories.

METHODS

Study Design

We used GBTMs to analyze longitudinal school administrative data across 5 timepoints retrospectively collected during a randomized trial of Advancement Via Individual Determination (AVID) (NCT03059433), a nationwide college readiness program that encourages enrollment in college preparatory classes and offers academic skills coaching and mentorship from teachers and college students.^23,24 The study was approved by the overseeing institutional review board and school district.

Study Context

Study sites were 5 public high schools in Los Angeles serving predominantly low-income, minoritized youth (Table 1). All students entering 9^th grade at a study high school with a middle school grade-point-average of 2.0–3.5 were eligible to participate in AVID. AVID eligibility and interest at study schools exceeded program capacity, therefore AVID enrollment occurred via a random admissions lottery which served as student-level randomization in the larger trial. AVID is perceived to influence adolescents’ academic and health outcomes in part by increasing exposure of middle-performing students to peers with higher academic performance and engagement and supportive adult mentors, as well as facilitating academic and social-emotional skill development needed to prepare for and persist in college.^23,24

Table 1.

Characteristics of Study Schools

Characteristic	School 1	School 2	School 3	School 4	School 5
Total enrollment	1504	1951	1030	2020	1650
% qualifying for free or reduced-price meals	90	91	90	85	64
% Hispanic	93	93	73	99	68
% Black	2	6	1	1	23

Participants

For the present study, participants were all students enrolled at study schools within the grade cohorts eligible for the larger trial regardless of students’ actual participation in the trial. This sample includes 2 cohorts of youth who completed 8^th grade in any district middle school and entered 9^th grade in a study high school in 2017 or 2018. Cohort 1 includes students in 8^th grade in 2016–2017 through 12^th grade in 2020–2021. Cohort 2 includes students in 8^th grade in 2017–2018 through 11^th grade in 2020–2021.

Data Collection

We collected data in partnership with the participating school district. School administrative data related to attendance, suspension, and AVID participation were recorded according to routine school district procedures for each academic year and deidentified before sharing with the research team.

Measures

School attendance.

We calculated attendance percentage by dividing total days attended by total days enrolled for each academic year and multiplying by 100. According to California Department of Education guidelines for reporting chronic absenteeism, we included only students enrolled for at least 31 days in the academic year.

Suspension.

Suspension was defined as receiving an out-of-school suspension in the academic year and included in GBTMs as a binary time-varying covariate. We hypothesized that suspension would have lasting negative effects on students’ attendance. All academic years prior to the first suspension had an indicator of 0 and all academic years including and following the first suspension had an indicator of 1.²⁵

AVID participation.

AVID participation was defined as course enrollment in the AVID academic skills elective in at least 1 semester and included in GBTMs as a binary time-varying covariate. We hypothesized that AVID would improve students’ attendance with the greatest effect when students were actively engaged in AVID programming. We modeled AVID participation to freely vary throughout the study period, with an indicator of 1 during academic years in which a student was enrolled in AVID.²⁵

Statistical Analysis

We used GBTMs to visualize group attendance trajectories, estimate the proportion belonging to each group, and estimate the influence of time-varying covariates trajectory shapes. Analyses were completed using the traj plugin for STATA v17 (StataCorp LLC, College Station, TX).

Our dependent variable was attendance in the 2016–2017 (Year 1), 2017–2018 (Year 2), 2018–2019 (Year 3), and 2020–2021 (Year 5) academic years. We excluded attendance in the 2019–2020 (Year 4) academic year as this coincided with initial COVID-19-related school closures and methods for tracking attendance were variable at this time. We used a staged procedure²⁵ to compare models with between 1–6 groups and all quadratic polynomial functions, compare models with polynomial functions varied from constant to quadratic, and adjust for time-stable covariates of study cohort and high school.⁸ We selected the optimal model at each stage based on: 1) highest Bayesian information criterion with log Bayes factor approximation ≥ 10; 2) highest relative entropy; 3) at least 5% of participants in each group; and 4) clinical meaningfulness of the trajectories.²⁵ We evaluated the fit of the final adjusted base model based on: 1) average posterior probabilities of group membership ≥ 0.7; 2) odds of correct classification ≥ 5.0; 3) similarity between the estimated probability and observed proportion of group membership; and 4) relatively constant and narrow confidence intervals.²⁵

Next, we estimated the influence of two school policies and practices on the identified attendance trajectories. We generated separate GBTMs including suspension and AVID participation as time-varying covariates in the final adjusted base model.²⁵ We used Wald tests to assess for differences in the effect of each covariate between attendance trajectory groups.

Finally, we conducted a post hoc sensitivity analysis. GBTMs account for data missing at random using maximum likelihood estimation.²⁵ Given the potential for nonignorable data missingness during COVID-19 era data collection, we summarized data missingness, excluded participants with any missing data, and repeated the model selection and assessment procedures above.

RESULTS

A total of 2,850 participants were included in this analysis including 1,698 in Cohort 1 and 1,152 in Cohort 2. Attendance data was available for 84.2% of possible data points. Data missingness varied by academic year and was 12.8% in 2016–2017, 10.0% in 2017–2018, 7.4% in 2018–2019, and 26.6% in 2020–2021. Forty-six participants (1.6%) experienced an out-of-school suspension during the study period, with only 5 participants (0.2%) experiencing more than 1 out-of-school suspension. Overall, 395 students (13.9%) participated in AVID during the study period.

Attendance Trajectories

We identified 3 attendance trajectory groups in the final adjusted base model which had good model fitness based on our criteria (Figure 1; Supplemental Tables 2–4). The stable high attendance group included 77.2% of the sample (n=2250) and displayed fairly constant high attendance throughout the study period. The acutely declining attendance group included 17.7% of the sample (n=455) and displayed initial high attendance with modest decline prior to COVID-19-related school closures during the 2019–2020 school year and more rapid decline after this time. The chronically declining attendance group included 5.1% of the sample (n=145) and displayed continual decline in attendance throughout.

Figure 1. Final adjusted base model including 3 attendance trajectory subgroups.

Solid lines represent estimated attendance trajectories for each group. Dashed lines represent 95% confidence intervals of the estimated probabilities of group membership. Points represent average observed values at each time point with responses weighted based on posterior probabilities of group membership.

The relationship between time-stable covariates included in the final adjusted base model and attendance trajectory group membership are detailed in Supplemental Table 5. Students in Cohort 2 were more likely to be members of the stable high attendance group as compared to the chronically declining attendance group (Odd Ratio=3.53; 95% CI 2.84–4.22). There were no other statistically significant relationships.

Effect of Time-varying Covariates

Visual representations of the estimated effect of suspension and AVID participation on attendance trajectories are displayed in Figure 2. Suspension was associated with declines in attendance across all groups (stable high attendance: β= −8.96; p < 0.0001; acutely declining attendance: β= −12.00; p < 0.0001; chronically declining attendance: β= −7.61; p = 0.025; Supplemental Table 6). These associations were not significantly different between groups (χ²= 1.57; p = 0.46). AVID participation was associated with slightly lower attendance in the stable high attendance group (β= −0.97; p = 0.025) and higher attendance in the acutely declining attendance group (β= 3.60; p = 0.015; Supplemental Table 7). Associations were significantly different between groups (χ²= 8.77; p = 0.013).

Figure 2. Effect of Suspension and AVID Participation in School Year 2 on Expected Attendance Trajectories:

A) Effect of an out-of-school suspension occurring in school year 2 on estimated school attendance trajectories; B) Effect of AVID participation beginning in school year 2 and continuing through school year 5 on estimated school attendance trajectories. In both panels, faded solid lines represent estimated attendance trajectories for each group in the final adjusted base model in the absence of suspension or AVID participation. Bold solid lines represent the shifted estimated attendance trajectories for each group after inclusion of suspension (A) or AVID participation (B) as time-varying covariates first occurring in the 2017–18 school year. Dashed lines represent 95% confidence intervals of the estimated probabilities of group membership. Points represent average observed values at each time point with responses weighted based on posterior probabilities of group membership. All models are adjusted for time-stable covariates of study cohort and high school.

Sensitivity Analysis

There were 1,696 participants (59.5%) with no missing data. Including only these participants, a 3-group model was again the best fitting. Most of the sample (78.6%) was again assigned to a group with stable high attendance (Group 1; Figure 3). 13.5% were assigned to a group with more gradual decline and greater average predicted attendance in Year 5 as compared to the acutely declining attendance group in the primary analysis (Group 2). 7.9% were assigned to a group with greater early average predicted attendance and more rapidly declining attendance following COVID-19-related school closures as compared to the chronically declining group in the primary analysis (Group 3).

Figure 3. Final adjusted base model identified in sensitivity analysis including only participants with all available attendance data.

Solid lines represent estimated attendance trajectories for each group. Dashed lines represent 95% confidence intervals of the estimated probabilities of group membership. Points represent average observed values at each time point with responses weighted based on posterior probabilities of group membership.

DISCUSSION

We used GBTMs to characterize the attendance patterns of rising high school students before and after the onset of the COVID-19 pandemic. We identified three distinct attendance trajectory groups: 1) stable high attendance, 2) acutely declining attendance, and 3) chronically declining attendance. Suspension negatively influenced each group. Participation in the AVID college readiness program had uniquely positive influence on attendance in the acutely declining attendance group.

Similar to prior research, this study revealed that attendance is often not stable over time and can vary in meaningful ways for different students.^19–21 Specifically, this study characterizes the range of attendance patterns underlying the increased rate of chronic absenteeism observed after the onset of COVID-19.¹⁷ Compared to one prior study characterizing high school attendance trajectories before the COVID-19 pandemic, our results display similarities and notable differences.¹⁹ Similarly, most students were assigned to trajectories characterized by good attendance throughout and fewer students were assigned to trajectories characterized by early good attendance with later progression to chronic absenteeism or early chronic absenteeism with continual worsening, corresponding to our stable high, acutely declining, and chronically declining attendance groups respectively.¹⁹ Uniquely, our two declining attendance trajectories displayed lower average predicted attendance in late high school and more rapidly declining attendance throughout, most apparent in the acutely declining attendance group following the onset of COVID-19.¹⁹ This suggests that although many students may experience minimal effects of the COVID-19 pandemic on their attendance, a sizeable minority may experience more severely declining attendance than expected throughout the high school years.

These findings have implications for adolescents’ educational outcomes and current and future health. Prior research suggests larger declines in attendance across the middle to high school transition may signify lower school connectedness and can increase the likelihood of shifting to a more problematic attendance trajectory over time.¹⁹ Although not a traditional school transition, the COVID-19 pandemic caused widespread disruptions to adolescents’ schooling and may function similarly with respect to attendance and school connectedness. Future research can determine if the effects of COVID-19 are transient or reshape attendance patterns longitudinally. In both research and clinical practice, a focus on individual-level or school-level point estimates of attendance may mask important variations in attendance trajectories over time and corresponding risk and protective factors which may be leveraged to improve outcomes. Early Warning Systems are already widely utilized tools that frequently use student-level indicators of attendance, school discipline, and course performance to help schools identify and intervene for students with elevated risk of failure to graduate.²⁶ Adapting such attendance monitoring protocols to include attendance data prior to and across the high school transition and to evaluate attendance relative to a student’s own baseline may identify declining attendance trajectories for early and tailored intervention.

In addition to individualized interventions, prior research has shown school structural characteristics including school size and teacher experience can influence students’ attendance trajectories.¹⁹ This study adds that policies and practices related to school discipline and college readiness programming may also influence attendance. Receiving any suspension was associated with worse attendance across all trajectories. This finding echoes prior work identifying enduring negative effects of receiving any out-of-school suspension on youth’s academic outcomes.²⁷ Exclusionary discipline leads directly to missed learning time and indirectly impacts attendance by disrupting school connections and reinforcing school disengagement.^3,22 Culturally responsive learning environments and restorative justice-based approaches can build connections instead of break them and promote positive youth development and school success for all.^22,28

The AVID program aims to identify students who may struggle to succeed when exposed to academically rigorous coursework and provide them support to increase their access to and preparation for college. Therefore, students eligible for AVID are likely facing more barriers to learning than those already succeeding academically. The unexpected finding that AVID was associated with a slight decline in attendance in the stable high attendance trajectory group may reflect this vulnerability. In contrast, students with declining attendance during high school may benefit from the enhanced support AVID provides, resulting in attendance decline that is not as steep.²³ The finding that AVID participation had a distinctly positive influence on attendance in the acutely declining attendance group—the group that appeared to suffer the most amplified negative effects of COVID—aligns with this theory. This both reinforces the utility of monitoring attendance patterns to guide tailored intervention and suggests that AVID or similar programs that surround youth with supportive adults and academically engaged peers may be important ways to restore or reroute education and health trajectories.³

This study has several limitations. The sample includes one large public school district in Southern California. The magnitude of predicted attendance rates and distribution of students across attendance trajectory groups may lack generalizability, but shapes of identified attendance trajectories likely hold insights to guide attendance monitoring and intervention in different settings. The use of school administrative data is a strength and limitation. School attendance data is more accurate than self-reported attendance and captures students with low attendance who may be excluded in other forms of school-based data collection.²⁹ However, we were limited to variables in administrative datasets and were unable to explore relevant covariates that may influence attendance including student demographics. Future research should explore associations between attendance and individual- (e.g., gender, race/ethnicity, primary language^20,21), school- (e.g., school socioeconomic status, teacher experience¹⁹), and structural-level (e.g., school segregation¹⁹) factors. We were also unable to differentiate between excused and unexcused absences. Although all absences contribute to poor educational outcomes, excused absences related to COVID-19 illness and isolation may have been high during the study resulting in lower overall attendance. The COVID-19 pandemic is a critical time to investigate attendance to understand and improve resulting education and health outcomes. However, questions remain as to the meaning and reliability of attendance data during periods of virtual instruction and the observed chronic absenteeism rate at the peak of the COVID-19 pandemic may be an underestimation.³⁰ If so, our identified attendance trajectories may underestimate the severity of attendance decline and the proportion of youth experiencing declining attendance. Also, attendance data not missing at random would not have been fully accounted for in our GBTMs. It is reassuring that our sensitivity analysis revealed the same number of groups, but some differences in the shapes of attendance trajectories were seen. We recommend that readers consider the GBTMs with and without missing data collectively while bearing in mind that the sample without missing data may exclude youth who transferred, were expelled, or withdrew from school during the study. Finally, few participants experienced more than 1 out-of-school suspension during the study period and data on alternate forms of school discipline (e.g., detention, in-school suspension) were unavailable. We were unable to estimate the effect of more numerous, prolonged, or severe disciplinary episodes, which have had compounding negative effects on students’ academic outcomes in prior research.²⁷

In conclusion, we identified three distinct attendance trajectories among adolescents enrolled in high school at the onset of the COVID-19 pandemic. Although many students maintained good attendance both before and after the onset of COVID-19, others experienced chronic or acute declines in attendance. Suspension was associated with worse attendance across trajectory groups, while participation in AVID was associated with improved attendance only for those with acutely declining attendance following onset of COVID-19. Monitoring school attendance trajectories, providing individualized intervention, and encouraging supportive school-level policies and practices are potentially important pathways to foster youth connectedness, enhance educational outcomes, and promote healthy futures.

Table 2.

Adjusted Base Model Fit Diagnostics

Group	Average Posterior Probability	Odds of Correct Classification	Estimated Probability of Subgroup Membership	Observed Subgroup Membership
Adjusted base model with 3 groups
Stable high attendance	0.957	6.00	0.789	0.772
Acutely declining attendance	0.884	40.12	0.160	0.177
Chronically declining attendance	0.931	251.73	0.051	0.051

Abbreviations:

COVID-19

Coronavirus disease 2019

AVID

Advancement Via Individual Determination

GBTM

group-based trajectory model